Skip to main content
Fig. 2 | Journal of NeuroEngineering and Rehabilitation

Fig. 2

From: Robust walking control of a lower limb rehabilitation exoskeleton coupled with a musculoskeletal model via deep reinforcement learning

Fig. 2

Overview of the modular, decoupled RL-based walking control framework of the LLRE with human-in-the-loop. The framework is separated into two parts: A muscle-actuated human policy training. B LLRE control policy training, which integrates three deep neural networks (marked with pink blocks): an RL-based interaction network for human control to manage human-exoskeleton interface forces; a supervised muscle coordination network for whole body muscle control; a RL-based motion imitation network for the control of the LLRE. The muscle-actuated human control loop A is designed to learn human muscle coordination giving the health status of the human and predicted exoskeleton assistance. The LLRE control loop B is designed to imitate a target walking motion while maintaining strong robustness and balance under the human-exoskeleton interaction. These three networks can be jointly trained in the simulation while they interact with each other to achieve maximum rewards during RL

Back to article page